Extending automorphisms of normal algebraic fields Matthew Harrison-Trainor University of California, Berkeley AMS Sectional Meeting, Charleston, SC, March 2017 This is joint work with Russell Miller and Alexander Melnikov. I will be talking about the effective versions of the following facts about fields: Every embedding of a field F into an algebraically closed field K extends to an embedding of F into K. Every automorphism of a field F extends to an automorphism of F . First we will review some effective field theory. Let F be a computable field. Definition The splitting set SF of F is the set of all polynomials p ∈ F [X ] which are reducible over F . If SF is computable, we say that F has a splitting algorithm. Theorem (Rabin's embedding theorem) There is a computable algebraically closed field F and a computable field embedding {∶ F → F such that F is algebraic over {(F ). For any such F and {, the image {(F ) of F in F is Turing equivalent to the splitting set of F . Theorem (Kronecker) If F has a splitting algorithm, then every finite extension of F has a splitting algorithm. We want to know: When does a computable embedding of a field F into an algebraically closed field K extend to a computable embedding of F into K? When does a computable automorphism of a field F extend to a computable automorphism of F ? Friedman, Simpson, and Smith, and Dorais, Hirst, and Shafer analyzed these questions using Reverse Mathematics. We can state their results in terms of effective algebra. For embeddings into algebraically closed fields: Theorem (Friedman-Simpson-Smith; Dorais-Hirst-Shafer) Let F be a computable field and let {∶ F → F be a computable embedding of F into its algebraic closure. If F has a splitting algorithm, every computable embedding of F into a computable algebraically closed field K extends to a computable embedding of F into K. Even if F does not have a splitting algorithm, every computable embedding of F into a computable algebraically closed field K extends to a low embedding of F into K. For extensions of automorphisms: Theorem (Friedman-Simpson-Smith; Dorais-Hirst-Shafer) Let F be a computable field and let {∶ F → F be a computable embedding of F into its algebraic closure. If F has a splitting algorithm, every computable automorphism of F extends to a computable automorphism of F . Even if F does not have a splitting algorithm, every computable automorphism of F extends to a low automorphism of F . We will try to answer the question: is it necessary to have a splitting algorithm? Theorem (HT-Miller-Melnikov) Let F be a computable field and let {∶ F → F be a computable embedding of F into its algebraic closure. The following are equivalent: 1 F has a splitting algorithm. 2 Every computable embedding of F into a computable algebraically closed field K extends to a computable embedding of F into K. β / FO ?K { α F Theorem (HT-Miller-Melnikov) Let F be a computable normal algebraic extension of the prime field and let {∶ F → F be a computable embedding of F into its algebraic closure. The following are equivalent: 1 F has a splitting algorithm. 2 Every computable automorphism of F extends to a computable automorphism of F . β / FO FO { { / F α F Before, we fixed the embedding of F into F . What happens if we let this embedding vary? Question Which fields F have the following property? For every computable automorphism α of F , there is a computable embedding {∶ F → F of F into an algebraic closure and a computable automorphism β of F extending α. We do not have a complete solution to this question, but towards a partial solution, we introduce the non-covering property. Definition We say that a group G has the non-covering property if for all finite index normal subgroups M ⊊ N of G and g ∈ G, there is h ∈ gN such that for all −1 x ∈ G, x hx ∉ gM. Lemma Let F ~E be a separable normal extension. The following are equivalent: 1 Gal(F ~E) has the non-covering property. 2 For all finite normal subextensions K1~E and K2~E with K2 ⊈ K1, and every pair of automorphisms σ of K1 and τ of K2 fixing E, there is an automorphism α of F extending σ and incompatible with τ (i.e., (K2; τ) does not embed into (F ; α) as a difference field). Theorem (HT-Miller-Melnikov) Let F be a computable normal algebraic extension of the prime field Fp such that Gal(F ~Fp) has the non-covering property. The following are equivalent: 1 F has a splitting algorithm. 2 For every computable automorphism α of F , there is a computable embedding {∶ F → F of F into an algebraic closure and a computable automorphism β of F extending α. β / FO FO { { / F α F The following groups have the non-covering property: abelian groups, simple groups, the quaternion group. S3 does not have the non-covering property. Theorem (HT-Miller-Melnikov) Let {Gi ∶ i ∈ I } be a collection of profinite groups, each of which has the non-covering property. Then ∏i∈I Gi has the non-covering property. Theorem (HT-Miller-Melnikov) Let F be a computable normal algebraic extension of Fp in characteristic p > 0. The following are equivalent: 1 F has a splitting algorithm. 2 For every computable automorphism α of F , there is a computable embedding {∶ F → F of F into an algebraic closure and a computable automorphism β of F extending α. Proof. The Galois group of every normal extension F ~Fp in characteristic p > 0 is abelian and hence has the non-covering property. Question Is this true in characteristic zero?.